TABLE 3.
Regression Coefficients and AIC for Visit-Level Modeling of Hypomaniaa
| Coefficient | Full Model | Clinical-Only Model | DBS-Only Model | Iterative Optimal Model |
|---|---|---|---|---|
| Intercept | 65.96 (7642.33) | −3.45 (1.22)* | 63.10 (8816.64) | −1.91 (0.66)* |
| Age | −0.06 (0.06) | −0.01 (0.03) | ||
| Male sex | −3.36 (1.79) | −0.78 (0.68) | −1.61 (0.68)** | |
| Left | ||||
| Contact | ||||
| 0 | −15.62 (1827.12) | −16.44 (2245.99) | ||
| 1 | −13.60 (1827.12) | −16.02 (2245.99) | ||
| 2 | −14.60 (1827.12) | −17.04 (2245.99) | −1.23 (0.87) | |
| 3 | −13.84 (1827.12) | −16.14 (2245.99) | ||
| Monopolar stimulation | −18.42 (1827.12) | −17.12 (2245.99) | ||
| Voltage | 0.33 (0.35) | 0.25 (0.25) | ||
| Frequency, 130 | −0.43 (1.58) | −1.03 (1.40) | ||
| Pulse width | −0.02 (0.02) | −0.01 (0.02) | ||
| Right | ||||
| Contact, right | ||||
| 0 | −20.99 (3356.03) | −19.22 (3793.26) | −2.89 (0.87)*** | |
| 1 | −18.41 (3356.03) | −17.87 (3793.26) | ||
| 2 | −16.11 (3356.03) | −17.08 (3793.26) | ||
| 3 | −18.88 (3356.03) | −16.86 (3793.26) | ||
| Monopolar stimulation | −1.09 (3356.03) | −15.41 (3793.26) | 1.56 (0.68)** | |
| Voltage | −0.39 (0.38) | −0.18 (0.25) | ||
| Frequency, 130 | −0.69 (1.99) | 0.26 (1.61) | ||
| Pulse width | 0.03 (0.02) | 0.01 (0.02) | ||
| Constant .4 weeks | 0.89 (1.04) | 0.75 (0.77) | ||
| MADRS slope | −1.00 (0.61) | −0.71 (0.33)** | ||
| OCD | 1.06 (2.03) | −0.10 (0.88) | ||
| History of mania | −0.73 (2.75) | 0.25 (1.11) | ||
| Family history of bipolar disorder | −0.24 (1.85) | 1.21 (0.83) | ||
| Acute mood change | 2.05 (2.16) | 1.15 (1.21) | ||
| Acute energy change | 0.31 (2.10) | 1.70 (1.22) | ||
| AIC | 117.51 | 117.22 | 113.88 | 93.11 |
| Log likelihood | −32.75 | −49.61 | −38.94 | −41.55 |
Coefficients are given as a maximum likelihood estimate (standard error of the mean). The full and DBS-only models contain multiple terms with very high standard errors, implying that the data do not support fitting of models with this many parameters. The clinical only model does not suffer from this problem but has poor predictive value (AUC=0.64 in Figure 3). The iteratively discovered model has reasonable parameter estimates with narrow standard errors of the mean while still yielding a good AUC of 0.84. AIC, Akaike’s information criterion; DBS, deep brain stimulation; MADRS, Montgomery-Åsberg Depression Rating Scale; OCD, obsessive-compulsive disorder.
p<0.01.
p<0.05.
p<0.001 (Wald z-test).